When amphibians thermoregulate, they face a fundamental trade-off between the ability to maintain activity and an increased rate of dehydration at higher temperatures. Canopy coverage affects both the thermal and hydric conditions of the environment and can therefore influence amphibian thermoregulation. Frogs require proper conditions to thermoregulate to successfully grow, survive, and reproduce. But while we know how canopy and environmental variables typically affect operative temperature, less is known about the effects on amphibian water loss rates. In this study, we measure the effect of canopy coverage on the conditions available for thermoregulation at a breeding pond of the California red-legged frog, Rana draytonii. We use agar frog models to estimate the thermal and hydric capacities frogs would experience in locations with different canopy coverage and microhabitats. At each site, we deployed models under four microhabitat treatments: wet/sun, wet/shade, dry/sun, and dry/shade. We modeled how environmental variables affected operative temperature and evaporative water loss from agar frogs. We found positive effects of air temperature, the sun treatment, and reduced canopy cover on operative temperature, and negative direct or indirect effects of these variables on evaporative water loss, consistent with the hypothesized trade-off between thermoregulatory behavior to increased temperature and the increased desiccation risk due to higher water loss. Additionally, our results indicate that the availability of wet microhabitats can allow frogs to reduce water loss, potentially mitigating the risk of desiccation when thermoregulating to achieve higher operative temperatures. Our findings suggest, that with access to proper microhabitats, amphibians can mitigate the fundamental trade-off and receive benefits of thermoregulating at high temperatures.

Study Design

To measure the thermoregulatory conditions at the pond, we divided up the pond into 8 wedges, each with an area of approximately 51.28m². We deployed agar frogs in each site’s shoreline for 24-hour periods from March to September. We used natural differences in canopy coverage to test how percent canopy coverage affects EWL and T_e on agar frogs. For a list of variables and their definitions, see Table 1.

To measure the full range of thermal and hydric conditions an amphibian could experience, agar frogs were placed in different microhabitats using a 2 x 2 factorial design with two levels of light treatment (sun vs shade) and two levels of water treatment (wet vs dry). We tracked T_e with internal dataloggers, EWL from mass changes in agar frogs, and other environmental conditions. With this design, we tested how percent canopy coverage affects agar frogs’ T_e and EWL individually. Then, we tested the bi-directional interaction between T_e and EWL using structural equation modeling. Finally, we tested a real-life example of thinning excess vegetation and its effect on EWL and T_e.

Canopy Coverage

Holding a camera at 1 meter height and pointing the lens up vertically during the daytime, we took photographs of the canopy overstory at each site to calculate the percent canopy coverage. We took two photos at each site: one at the densest canopy coverage where shade agar frogs were deployed and the second at the thinnest canopy coverage where sun agar frogs were deployed (approximately 5m apart). We analyzed photos in Adobe Photoshop software and converted pixels of the sky to white and pixels of vegetation to black following the procedure in Stewart et al. 2007. Using ImageJ software (Rasband 1997-2018), we calculated the number of black vs white pixels estimated percent canopy coverage by comparing the number of pixels of vegetation to the total pixels. The average percent canopy coverage between the two photos was calculated for each site.

Agar Creation and Validation

Agar frogs are regularly used to estimate the T_e and EWL an amphibian could potentially access in its environment. The free evaporation from agar captures the evaporative properties experienced by live frogs, including the cooling aspects caused by EWL, since we are unable to use live frogs due to their threatened status. Agar EWL has been validated in comparison to living frogs, and agar models exhibit internal temperatures that match live amphibians. While there has been recent debate about using agar models to compare to live thermoregulating amphibians, we minimized identified limitations by using agar frogs that represent live frogs in size, shape, posture, and inactivity, and replace models facing extreme dehydration to avoid inaccurate readings of EWL. We also acknowledge that agar frogs cannot replicate live frog behavior and movements, so we instead use agar frogs to study the available microhabitats with different thermal and hydric opportunities for frogs, as opposed to simulating what a live frog would experience.

We made agar frog models in latex molds, which were initially created with plaster casts from museum specimens spanning the size range of anuran in the Pacific Northwest: two large-bodied frogs (15-48 grams), Bufo boreas and Rana draytonii, one small frog (6-22 grams), Pseudacris regilla, and a small ellipse shape (3 cm diameter, 5-16 grams). We filled molds with agar (4.74 g in 100 ml) tinted green with acrylic paint and allowed them to solidify around a thermocouple probe in the center. The agar formula we used was denser than previous studies to achieve a similar density to live amphibians, obtain more accurate T_e, and be a better proxy for simulating evaporative properties.

Agar Frog Deployment

We deployed the agar frogs for 24 hours every two weeks during the 2017 breeding season March-September (N = 13 total weeks of deployment). For the first 7 deployment weeks, agar frogs were deployed at all 8 sites. However, as the pond water levels naturally went down, agar was only deployed at sites still holding water, with four sites measured in weeks 8 and 9, three sites in weeks 10-12, and one site in week 13. Within each of the 8 wedges, one small (5–22 g) and one large (15–48 g) agar frog were deployed under four microhabitat treatments: wet/sun (N = 432 deployment periods), wet/shade (N = 438), dry/sun (N = 428), and dry/shade (N = 448). Agar frogs in the sunlight treatment were placed in the site under the location of lowest canopy coverage and shade light treatment frogs were placed under fallen branches to be fully shaded. Agar frogs in the dry water treatment were placed on a dry substrate near the pond bank (1-2 meters from the water’s edge), and wet water treatment frogs were placed on the water line, so half the agar was in the water and half was on the saturated substrate. As the season progressed and the water level changed, we adjusted the wet treatment location to keep agar frogs on the water line, but dry treatment locations remained in the original location. Every agar frog was deployed with an internal thermocouple (Onset HOBO ProV2 U23-003) that measured internal model temperature at one-minute intervals to calculate T_e.

To determine EWL for each deployment period, agar frogs were initially weighed with a Pesola PPS200 digital scale at each site location. Agar frogs were then deployed in the evening (15:00-19:30 hours) and reweighed three times in situ the following day: morning (between 07:30-11:00 hours), noon (between 11:00-15:00 hours) and evening (between 14:00-20:00 hours). This creates three deployment periods each day (overnight, morning, and afternoon), defined as the 4–14 hour period between models being set in the field and the time when the next weighing occurred (weigh point). The first 15 minutes after each weigh point were removed from calculations to ensure agar frogs had achieved equilibrium again following handling. Any agar frogs that had lost 10% or more of their body weight were replaced with fresh agar frogs because agar that has lost more than 10% mass no longer maintains consistent EWL with live amphibians. Cages, constructed from chicken wire, were placed over agar frogs to prevent damage from birds. A relative humidity datalogger (Onset HOBO ProV2 U23-001) was placed at the main site 2 meters above the ground in shade conditions to log ambient air temperature and relative humidity at one-minute intervals.

EWL and T_e Calculations

EWL was measured as percent water loss per hour, calculated as the change in agar mass over the deployment period. Operative temperature, T_e, was calculated from the internal thermocouple data. We calculated the average, minimum, and maximum temperature for each weigh period as well as the standard deviation and full range difference of temperatures. The ambient data logger was used to calculate air temperature and relative humidity for each weight period. We calculated the average, minimum, maximum, standard deviation, and range of both air temperature and relative humidity.

Effects of Vegetation Trimming

At one wedge site along the pond, site managers trimmed the canopy and removed dead vegetation to simulate historical management. We used this site in our study, but we additionally compared how this artificially trimmed canopy affected EWL and T _e, compared to sites with naturally thin canopy coverage. We aimed to test if artificial thinning can return potential thermoregulatory benefits, such as increased T_e, that naturally thin canopy provides.

Statistical Analysis

Environmental Variable Effects on Average T_e and EWL

To examine how percent canopy coverage, water treatment, and light treatment impacted agar frogs, we fit a linear mixed model using the lmer package in R ver 3.6.1 (www.r-project.org). This modeling framework allowed us to examine the effect of many environmental variables (i.e., fixed effects) on multiple agar response variables and include two nested random effects of “session/period” and “site/microhabitat." We created two models to test how these environmental variables influenced two primary response variables: 1) EWL and 2) average T_e.

We started with a base model that includes the nested random effects above and the planned experimental variables as fixed effects: percent canopy coverage, water treatment, and light treatment. We chose these variables to start because percent canopy coverage is the predictor variable of interest, and the water and light treatments were manipulated in our experimental design. These nested effects and three fixed effects stayed constant in each following step of model creation.

Next, we added additional predictor variables using a traditional forward selection procedure, in each step adding the predictor that contributed most to the model (lowest p-value). Predictors of interest used in model selection were relative humidity, air temperature, day condition, agar mass, agar surface area, and agar shape. Additionally, we used the predictor of T_e for the EWL model and EWL for the T_emodel to start understanding the causal direction of the relationship between T_e and EWL. We did not use any interaction terms. We continued to add predictors and stopped when no predictors significantly improved the model (p>0.05). The significance of the fixed effect was assessed by using an F test in the ANOVA function in R, by sequentially adding the variable of interest and comparing it with the previous model, which did not include the effect. To reduce problems with multicollinearity, we only included the best single predictor variable (average, minimum, maximum, standard deviation, or range) in each of the following categories of highly correlated predictors: air temperature, agar operative temperature, and relative humidity. After one predictor was added from each category, other variables from that category were removed from consideration as future predictors.

We visually assessed the model assumptions of normality and homoscedasticity with Q-Q plots and plotted the model residuals against the fitted values. We also measured normality with a Shapiro-Wilk statistic test. We calculated the variance inflation factor to check for multicollinearity between predictor variables in the final models and used a variance inflation factor of less than 5 as our threshold for variable inclusion.

Interaction between T_e and EWL

To evaluate hypothesized causal pathways linking the EWL and average T_e of the agar frogs, including a potentially bidirectional relationship between T_e and EWL, we used piecewise structural equation modeling (SEM). This statistical technique estimates the strengths of the effects using the linear relationships between variables and can account for nested random effects. Structural equation modeling has been used as a strong technique to discern the bidirectional effects of two biological variables, which are each individually affected by additional variables. For discrete variables, such as light treatment and water treatment, we set them as binary values (0 and 1) to model as numeric variables. The hypothesized causal links between variables were determined by the LMMs created in the previous section for EWL and average T_e. To fit our models, we used the piecewiseSEM package in R.

Effects of Vegetative Trimming

To discern the effects of the artificial trimming on EWL and T_e, we used a 1 sample t-test. With this test, we tested how the distribution of EWL and T_e of the artificially trimmed site compared to the predicted values of our model.